net/netmap.c

   1 /*
   2  * netmap access for qemu
   3  *
   4  * Copyright (c) 2012-2013 Luigi Rizzo
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25
  26 #include <sys/ioctl.h>
  27 #include <net/if.h>
  28 #include <sys/mman.h>
  29 #include <stdint.h>
  30 #include <stdio.h>
  31 #define NETMAP_WITH_LIBS
  32 #include <net/netmap.h>
  33 #include <net/netmap_user.h>
  34
  35 #include "net/net.h"
  36 #include "net/tap.h"
  37 #include "clients.h"
  38 #include "sysemu/sysemu.h"
  39 #include "qemu/error-report.h"
  40 #include "qemu/iov.h"
  41
  42 /* Private netmap device info. */
  43 typedef struct NetmapPriv {
  44     int                 fd;
  45     size_t              memsize;
  46     void                *mem;
  47     struct netmap_if    *nifp;
  48     struct netmap_ring  *rx;
  49     struct netmap_ring  *tx;
  50     char                fdname[PATH_MAX];        /* Normally "/dev/netmap". */
  51     char                ifname[IFNAMSIZ];
  52 } NetmapPriv;
  53
  54 typedef struct NetmapState {
  55     NetClientState      nc;
  56     NetmapPriv          me;
  57     bool                read_poll;
  58     bool                write_poll;
  59     struct iovec        iov[IOV_MAX];
  60     int                 vnet_hdr_len;  /* Current virtio-net header length. */
  61 } NetmapState;
  62
  63 #ifndef __FreeBSD__
  64 #define pkt_copy bcopy
  65 #else
  66 /* A fast copy routine only for multiples of 64 bytes, non overlapped. */
  67 static inline void
  68 pkt_copy(const void *_src, void *_dst, int l)
  69 {
  70     const uint64_t *src = _src;
  71     uint64_t *dst = _dst;
  72     if (unlikely(l >= 1024)) {
  73         bcopy(src, dst, l);
  74         return;
  75     }
  76     for (; l > 0; l -= 64) {
  77         *dst++ = *src++;
  78         *dst++ = *src++;
  79         *dst++ = *src++;
  80         *dst++ = *src++;
  81         *dst++ = *src++;
  82         *dst++ = *src++;
  83         *dst++ = *src++;
  84         *dst++ = *src++;
  85     }
  86 }
  87 #endif /* __FreeBSD__ */
  88
  89 /*
  90  * Open a netmap device. We assume there is only one queue
  91  * (which is the case for the VALE bridge).
  92  */
  93 static int netmap_open(NetmapPriv *me)
  94 {
  95     int fd;
  96     int err;
  97     size_t l;
  98     struct nmreq req;
  99
 100     me->fd = fd = open(me->fdname, O_RDWR);
 101     if (fd < 0) {
 102         error_report("Unable to open netmap device '%s' (%s)",
 103                         me->fdname, strerror(errno));
 104         return -1;
 105     }
 106     memset(&req, 0, sizeof(req));
 107     pstrcpy(req.nr_name, sizeof(req.nr_name), me->ifname);
 108     req.nr_ringid = NETMAP_NO_TX_POLL;
 109     req.nr_version = NETMAP_API;
 110     err = ioctl(fd, NIOCREGIF, &req);
 111     if (err) {
 112         error_report("Unable to register %s: %s", me->ifname, strerror(errno));
 113         goto error;
 114     }
 115     l = me->memsize = req.nr_memsize;
 116
 117     me->mem = mmap(0, l, PROT_WRITE | PROT_READ, MAP_SHARED, fd, 0);
 118     if (me->mem == MAP_FAILED) {
 119         error_report("Unable to mmap netmap shared memory: %s",
 120                         strerror(errno));
 121         me->mem = NULL;
 122         goto error;
 123     }
 124
 125     me->nifp = NETMAP_IF(me->mem, req.nr_offset);
 126     me->tx = NETMAP_TXRING(me->nifp, 0);
 127     me->rx = NETMAP_RXRING(me->nifp, 0);
 128     return 0;
 129
 130 error:
 131     close(me->fd);
 132     return -1;
 133 }
 134
 135 /* Tell the event-loop if the netmap backend can send packets
 136    to the frontend. */
 137 static int netmap_can_send(void *opaque)
 138 {
 139     NetmapState *s = opaque;
 140
 141     return qemu_can_send_packet(&s->nc);
 142 }
 143
 144 static void netmap_send(void *opaque);
 145 static void netmap_writable(void *opaque);
 146
 147 /* Set the event-loop handlers for the netmap backend. */
 148 static void netmap_update_fd_handler(NetmapState *s)
 149 {
 150     qemu_set_fd_handler2(s->me.fd,
 151                          s->read_poll  ? netmap_can_send : NULL,
 152                          s->read_poll  ? netmap_send     : NULL,
 153                          s->write_poll ? netmap_writable : NULL,
 154                          s);
 155 }
 156
 157 /* Update the read handler. */
 158 static void netmap_read_poll(NetmapState *s, bool enable)
 159 {
 160     if (s->read_poll != enable) { /* Do nothing if not changed. */
 161         s->read_poll = enable;
 162         netmap_update_fd_handler(s);
 163     }
 164 }
 165
 166 /* Update the write handler. */
 167 static void netmap_write_poll(NetmapState *s, bool enable)
 168 {
 169     if (s->write_poll != enable) {
 170         s->write_poll = enable;
 171         netmap_update_fd_handler(s);
 172     }
 173 }
 174
 175 static void netmap_poll(NetClientState *nc, bool enable)
 176 {
 177     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 178
 179     if (s->read_poll != enable || s->write_poll != enable) {
 180         s->read_poll = enable;
 181         s->read_poll = enable;
 182         netmap_update_fd_handler(s);
 183     }
 184 }
 185
 186 /*
 187  * The fd_write() callback, invoked if the fd is marked as
 188  * writable after a poll. Unregister the handler and flush any
 189  * buffered packets.
 190  */
 191 static void netmap_writable(void *opaque)
 192 {
 193     NetmapState *s = opaque;
 194
 195     netmap_write_poll(s, false);
 196     qemu_flush_queued_packets(&s->nc);
 197 }
 198
 199 static ssize_t netmap_receive(NetClientState *nc,
 200       const uint8_t *buf, size_t size)
 201 {
 202     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 203     struct netmap_ring *ring = s->me.tx;
 204     uint32_t i;
 205     uint32_t idx;
 206     uint8_t *dst;
 207
 208     if (unlikely(!ring)) {
 209         /* Drop. */
 210         return size;
 211     }
 212
 213     if (unlikely(size > ring->nr_buf_size)) {
 214         RD(5, "[netmap_receive] drop packet of size %d > %d\n",
 215                                     (int)size, ring->nr_buf_size);
 216         return size;
 217     }
 218
 219     if (nm_ring_empty(ring)) {
 220         /* No available slots in the netmap TX ring. */
 221         netmap_write_poll(s, true);
 222         return 0;
 223     }
 224
 225     i = ring->cur;
 226     idx = ring->slot[i].buf_idx;
 227     dst = (uint8_t *)NETMAP_BUF(ring, idx);
 228
 229     ring->slot[i].len = size;
 230     ring->slot[i].flags = 0;
 231     pkt_copy(buf, dst, size);
 232     ring->cur = ring->head = nm_ring_next(ring, i);
 233     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 234
 235     return size;
 236 }
 237
 238 static ssize_t netmap_receive_iov(NetClientState *nc,
 239                     const struct iovec *iov, int iovcnt)
 240 {
 241     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 242     struct netmap_ring *ring = s->me.tx;
 243     uint32_t last;
 244     uint32_t idx;
 245     uint8_t *dst;
 246     int j;
 247     uint32_t i;
 248
 249     if (unlikely(!ring)) {
 250         /* Drop the packet. */
 251         return iov_size(iov, iovcnt);
 252     }
 253
 254     last = i = ring->cur;
 255
 256     if (nm_ring_space(ring) < iovcnt) {
 257         /* Not enough netmap slots. */
 258         netmap_write_poll(s, true);
 259         return 0;
 260     }
 261
 262     for (j = 0; j < iovcnt; j++) {
 263         int iov_frag_size = iov[j].iov_len;
 264         int offset = 0;
 265         int nm_frag_size;
 266
 267         /* Split each iovec fragment over more netmap slots, if
 268            necessary. */
 269         while (iov_frag_size) {
 270             nm_frag_size = MIN(iov_frag_size, ring->nr_buf_size);
 271
 272             if (unlikely(nm_ring_empty(ring))) {
 273                 /* We run out of netmap slots while splitting the
 274                    iovec fragments. */
 275                 netmap_write_poll(s, true);
 276                 return 0;
 277             }
 278
 279             idx = ring->slot[i].buf_idx;
 280             dst = (uint8_t *)NETMAP_BUF(ring, idx);
 281
 282             ring->slot[i].len = nm_frag_size;
 283             ring->slot[i].flags = NS_MOREFRAG;
 284             pkt_copy(iov[j].iov_base + offset, dst, nm_frag_size);
 285
 286             last = i;
 287             i = nm_ring_next(ring, i);
 288
 289             offset += nm_frag_size;
 290             iov_frag_size -= nm_frag_size;
 291         }
 292     }
 293     /* The last slot must not have NS_MOREFRAG set. */
 294     ring->slot[last].flags &= ~NS_MOREFRAG;
 295
 296     /* Now update ring->cur and ring->head. */
 297     ring->cur = ring->head = i;
 298
 299     ioctl(s->me.fd, NIOCTXSYNC, NULL);
 300
 301     return iov_size(iov, iovcnt);
 302 }
 303
 304 /* Complete a previous send (backend --> guest) and enable the
 305    fd_read callback. */
 306 static void netmap_send_completed(NetClientState *nc, ssize_t len)
 307 {
 308     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 309
 310     netmap_read_poll(s, true);
 311 }
 312
 313 static void netmap_send(void *opaque)
 314 {
 315     NetmapState *s = opaque;
 316     struct netmap_ring *ring = s->me.rx;
 317
 318     /* Keep sending while there are available packets into the netmap
 319        RX ring and the forwarding path towards the peer is open. */
 320     while (!nm_ring_empty(ring) && qemu_can_send_packet(&s->nc)) {
 321         uint32_t i;
 322         uint32_t idx;
 323         bool morefrag;
 324         int iovcnt = 0;
 325         int iovsize;
 326
 327         do {
 328             i = ring->cur;
 329             idx = ring->slot[i].buf_idx;
 330             morefrag = (ring->slot[i].flags & NS_MOREFRAG);
 331             s->iov[iovcnt].iov_base = (u_char *)NETMAP_BUF(ring, idx);
 332             s->iov[iovcnt].iov_len = ring->slot[i].len;
 333             iovcnt++;
 334
 335             ring->cur = ring->head = nm_ring_next(ring, i);
 336         } while (!nm_ring_empty(ring) && morefrag);
 337
 338         if (unlikely(nm_ring_empty(ring) && morefrag)) {
 339             RD(5, "[netmap_send] ran out of slots, with a pending"
 340                    "incomplete packet\n");
 341         }
 342
 343         iovsize = qemu_sendv_packet_async(&s->nc, s->iov, iovcnt,
 344                                             netmap_send_completed);
 345
 346         if (iovsize == 0) {
 347             /* The peer does not receive anymore. Packet is queued, stop
 348              * reading from the backend until netmap_send_completed()
 349              */
 350             netmap_read_poll(s, false);
 351             break;
 352         }
 353     }
 354 }
 355
 356 /* Flush and close. */
 357 static void netmap_cleanup(NetClientState *nc)
 358 {
 359     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 360
 361     qemu_purge_queued_packets(nc);
 362
 363     netmap_poll(nc, false);
 364     munmap(s->me.mem, s->me.memsize);
 365     close(s->me.fd);
 366
 367     s->me.fd = -1;
 368 }
 369
 370 /* Offloading manipulation support callbacks. */
 371 static bool netmap_has_ufo(NetClientState *nc)
 372 {
 373     return true;
 374 }
 375
 376 static bool netmap_has_vnet_hdr(NetClientState *nc)
 377 {
 378     return true;
 379 }
 380
 381 static bool netmap_has_vnet_hdr_len(NetClientState *nc, int len)
 382 {
 383     return len == 0 || len == sizeof(struct virtio_net_hdr) ||
 384                 len == sizeof(struct virtio_net_hdr_mrg_rxbuf);
 385 }
 386
 387 static void netmap_using_vnet_hdr(NetClientState *nc, bool enable)
 388 {
 389 }
 390
 391 static void netmap_set_vnet_hdr_len(NetClientState *nc, int len)
 392 {
 393     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 394     int err;
 395     struct nmreq req;
 396
 397     /* Issue a NETMAP_BDG_VNET_HDR command to change the virtio-net header
 398      * length for the netmap adapter associated to 'me->ifname'.
 399      */
 400     memset(&req, 0, sizeof(req));
 401     pstrcpy(req.nr_name, sizeof(req.nr_name), s->me.ifname);
 402     req.nr_version = NETMAP_API;
 403     req.nr_cmd = NETMAP_BDG_VNET_HDR;
 404     req.nr_arg1 = len;
 405     err = ioctl(s->me.fd, NIOCREGIF, &req);
 406     if (err) {
 407         error_report("Unable to execute NETMAP_BDG_VNET_HDR on %s: %s",
 408                      s->me.ifname, strerror(errno));
 409     } else {
 410         /* Keep track of the current length. */
 411         s->vnet_hdr_len = len;
 412     }
 413 }
 414
 415 static void netmap_set_offload(NetClientState *nc, int csum, int tso4, int tso6,
 416                                int ecn, int ufo)
 417 {
 418     NetmapState *s = DO_UPCAST(NetmapState, nc, nc);
 419
 420     /* Setting a virtio-net header length greater than zero automatically
 421      * enables the offloadings.
 422      */
 423     if (!s->vnet_hdr_len) {
 424         netmap_set_vnet_hdr_len(nc, sizeof(struct virtio_net_hdr));
 425     }
 426 }
 427
 428 /* NetClientInfo methods */
 429 static NetClientInfo net_netmap_info = {
 430     .type = NET_CLIENT_OPTIONS_KIND_NETMAP,
 431     .size = sizeof(NetmapState),
 432     .receive = netmap_receive,
 433     .receive_iov = netmap_receive_iov,
 434     .poll = netmap_poll,
 435     .cleanup = netmap_cleanup,
 436     .has_ufo = netmap_has_ufo,
 437     .has_vnet_hdr = netmap_has_vnet_hdr,
 438     .has_vnet_hdr_len = netmap_has_vnet_hdr_len,
 439     .using_vnet_hdr = netmap_using_vnet_hdr,
 440     .set_offload = netmap_set_offload,
 441     .set_vnet_hdr_len = netmap_set_vnet_hdr_len,
 442 };
 443
 444 /* The exported init function
 445  *
 446  * ... -net netmap,ifname="..."
 447  */
 448 int net_init_netmap(const NetClientOptions *opts,
 449         const char *name, NetClientState *peer)
 450 {
 451     const NetdevNetmapOptions *netmap_opts = opts->netmap;
 452     NetClientState *nc;
 453     NetmapPriv me;
 454     NetmapState *s;
 455
 456     pstrcpy(me.fdname, sizeof(me.fdname),
 457         netmap_opts->has_devname ? netmap_opts->devname : "/dev/netmap");
 458     /* Set default name for the port if not supplied. */
 459     pstrcpy(me.ifname, sizeof(me.ifname), netmap_opts->ifname);
 460     if (netmap_open(&me)) {
 461         return -1;
 462     }
 463     /* Create the object. */
 464     nc = qemu_new_net_client(&net_netmap_info, peer, "netmap", name);
 465     s = DO_UPCAST(NetmapState, nc, nc);
 466     s->me = me;
 467     s->vnet_hdr_len = 0;
 468     netmap_read_poll(s, true); /* Initially only poll for reads. */
 469
 470     return 0;
 471 }
 472